Way more usage of tokens due to reasoning
What i personally feel, although the 3.5 flash model seems capable, the usage is linked across gemini models (flash/pro are ONE now!). Since the reasoning is a lot and it creates so many small inputs, it is burning way faster then regular gemini 3 flash used to. Literally gemini 3 flash was UNLIMITED, never reached the usage completely before.
Also, although they have improved the cooldown period of premium models of claude, etc, which is 5 hrs now, again the number of conversations in selfthought in a request eats away at your allowance and nothing is practically achieved by using opus. Before i could refactor major deficiencies in the app by using a single session, but now, it could not even get past formulating implementation plan.
I need to try antigravity CLI and checkout if less tokens are used.